Pith. sign in

REVIEW 2 major objections 5 minor 199 references

InSight-doc: Agentic Visual Perception for Long-Document Understanding

T0 review · 2 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A model that reads documents by zooming in beats fixed high-resolution reading.

desk verdict Real gains on document VQA from region-level adaptive zooming, but the missing dedup audit between the arXiv training snapshot and MMLongBench-Doc is the thing to chase first. read the letter →

arxiv 2608.10628 v1 pith:XNCSBJ22 submitted 2026-08-11 cs.CV cs.CLcs.LG

Kaican Li , Weiyan Xie , Lewei Yao , Jiannan Wu , Lanqing Hong , Yongxiang Huang , Nevin L. Zhang This is my paper · ORCID
classification cs.CVcs.CLcs.LG
keywords agenticvisualperceptionlong-documentunderstandingdocumentquestionansweringadaptiveresolutionzoom-inpolicyreinforcementlearninghallucinationreductionmultimodallargelanguagemodels
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

InSight-doc is an agentic visual perception method that lets a multimodal language model read long documents at low resolution first, then zoom into specific page regions when it needs finer evidence. The paper claims this turns visual resolution into an adaptive reasoning-time resource, allowing the model to match or beat fixed high-resolution processing while using many fewer tokens and less time. On document VQA benchmarks, the trained model improves accuracy by 4.3 to 16.4 points over its base model, cuts hallucination on unanswerable questions by more than 40 percent, and reduces inference latency by 41 to 68 percent. The core bet is that a model can learn where and when to zoom well enough to replace both external retrievers and full-page high-resolution encoding.

What carries the argument

The load-bearing mechanism is the zoom-in tool call, formalized as zoom_in(k, d, b | context), where k indexes the page image, d is a natural-language description of the region of interest, and b is a bounding box; the crop is taken from the high-resolution source and appended to the visual context. The learned policy decides when to call it, which region to request, and how to integrate the returned crop. Training uses supervised fine-tuning on 17.9K trajectories generated by a two-agent teacher, followed by GRPO reinforcement learning on 19.2K hard examples with binary accuracy reward. A latency bound shows that aggressive downsampling plus a few zoom calls keeps total inference below the no-resize baseline.

What would settle it

Retrain InSight-doc on the same SFT and RL data but with every teacher bounding box randomly shifted by roughly ten percent of page width, then evaluate on LongDocURL, which has box-level evidence annotations. If accuracy and evidence-box coverage stay close to the reported numbers, the model is not relying on precise teacher localization; if they collapse toward the baseline, the adaptive-perception claim depends on memorized teacher boxes.

Watch

Extended reading notes

Core claim

The central claim is that one end-to-end multimodal model can learn to acquire the right visual evidence at the right resolution through multi-round region cropping. Starting from a low-resolution overview of the whole document, the model emits a thought and a zoom-in tool call, receives a higher-resolution crop of a predicted bounding box, and repeats until it answers or abstains. The authors show this policy beats both the same backbone equipped with a zoom tool but no specialized training and the no-tool high-resolution baseline, while shortening sequence length and latency. It also abstains more accurately on unanswerable questions, which they attribute to the model actively checking regions, finding no evidence, and refusing to fabricate an answer.

Load-bearing premise

The teacher that produced the zoom-in training data must localize the right regions correctly enough; if its bounding boxes are systematically wrong or biased, the student inherits that bias and the claimed gains may not generalize beyond the training distribution.

Editorial extensions

If this is right

  • Documents can be processed at very low initial resolution without sacrificing accuracy, because the model recovers detail on demand.
  • Unanswerable questions become safer: the model more often states that evidence is insufficient instead of hallucinating an answer.
  • The learned zoom policy transfers from documents to general high-resolution image VQA, indicating a general visual-search ability.
  • Reinforcement learning sharpens the policy beyond supervised fine-tuning, raising evidence-box coverage and nearly eliminating stuck trajectories.
  • Because the method needs no external retriever, it avoids retrieval indexing overhead and retriever errors.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same zoom-as-resource idea could be applied to long videos, replacing spatial region cropping with temporal segment selection; the paper draws the analogy but does not test it.
  • If the teacher's localization quality is the bottleneck, then improving or scaling the teacher should directly improve the student, pointing to a distillation-style research program.
  • The token and latency savings could make long-document understanding feasible on smaller or edge-deployed models that cannot afford full-page high-resolution encoding.
  • A fully controlled head-to-head against retrieval-assisted page-level methods would settle whether region-level zooming is inherently more token-efficient; the paper's proxy comparison suggests so but is not conclusive.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper proposes InSight-doc, an agentic visual perception framework for long-document VQA in which the model starts from a low-resolution rendering of all pages and iteratively issues zoom_in tool calls (image index, region label, bounding box) to append high-resolution crops to its interleaved reasoning chain. The training corpus is built from six document sources, with an arXiv-derived branch that uses MinerU and Gemini to generate enriched visual descriptions and QA pairs, and an InSight-o3 two-agent teacher that produces zoom-in trajectories; this yields 17,913 SFT trajectories and 19,236 RL prompts. Starting from Qwen3-VL-8B-Instruct, the authors train with SFT then GRPO. On DUDE, MP-DocVQA, MMLongBench-Doc, and LongDocURL, the final model improves the baseline by 4.3-16.4 accuracy points, reduces hallucination on unanswerable questions by more than 40%, and reduces latency by 41%-68%. Additional experiments cover general high-resolution VQA (MME-RealWorld-Lite, O3-Bench), cross-paper comparisons, trajectory-quality metrics, an uncapped long-document setting, and an inference-cost analysis in Propositions 1 and 2.

Significance. If the results hold, this is a valuable empirical contribution: it demonstrates that a retriever-free, region-level zoom agent trained with SFT+RL can push the accuracy-efficiency Pareto frontier for long-document VQA. The controlled comparisons against Qwen3-VL-8B with and without the same zoom tool, the disclosure of the 40-page cap with uncapped results in Appendix F.2, the explicit caveats on cross-paper comparisons in Appendix F.3, the 150-example judge calibration, and the release of code, data, and model are all strengths. The main reservation is that the arXiv-derived training corpus may overlap with MMLongBench-Doc, and no deduplication is reported; since MMLongBench-Doc anchors the long-document part of the central claim, this issue must be resolved. A secondary concern is that the teacher trajectories are audited only on training data, leaving the localization-bias premise unverified on evaluation documents.

major comments (2)
  1. [Section 4.1 / Appendix C.1 / Appendix E.1 / Table 2] The training corpus uses an arXiv Kaggle snapshot that prioritizes longer papers with dense figures and tables (Section 4.1, Appendix C.1). MMLongBench-Doc is built from scientific papers with visualizations and contains documents up to 468 pages (Appendix E.1). The paper does not report any overlap or deduplication analysis between the arXiv training snapshot and MMLongBench-Doc or LongDocURL. Because Table 2's headline long-document gain is +17.1 points on MMLongBench-Doc at r=0.25, and the uncapped setting in Table 16 still shows a +17.0-point gap, partial overlap would inflate the central long-document claim: InSight-doc would have seen these document pages and their enriched descriptions during training, while the Qwen3-VL-8B baseline has not. Please report the overlap count (e.g., arXiv IDs or PDF hashes) against both long-document benchmarks and re-run the headline metrics on the non-overlapping subset (or exclude overlapping documents). This is necessary to support the long-document conclusions.
  2. [Section 4.2 / Appendix C.4 / Appendix D.2] The SFT trajectories are distilled from InSight-o3, in which a fine-tuned Qwen3-VL-8B vSearcher supplies the bounding box for every zoom-in call (Section 4.2, Appendix C.4). The trajectory-quality audit in Appendix D.2 is performed on training rows only, not on evaluation documents, so the transfer of the learned zoom policy depends on an unverified assumption that the teacher's localization is accurate and unbiased outside the training distribution. Table 6 reports evidence-box coverage for the final InSight-doc model, not for the teacher. Please add a small human audit of teacher bounding boxes on held-out evaluation-style documents, or report teacher-level coverage on a sample of LongDocURL evidence boxes, to show that the student is not inheriting systematic localization bias from the teacher.
minor comments (5)
  1. [Section 4.2 vs. Appendix C.4] The main text describes a universal three-stage cascade, but Appendix C.4 explains that arXiv-derived answerable rows bypass the Stage-1 prior-only filtering. This exception should be stated in Section 4.2.
  2. [Table 5 / Section 5.2 / Appendix F.3] The InSight-doc scores quoted in Table 5 for MMLongBench-Doc and LongDocURL (57.8 and 65.6 at r=0.5) correspond to the uncapped setting reported in Table 16, but Table 5's caption and the main-text discussion do not say so; add a footnote to prevent readers from comparing them with the capped numbers in Table 2.
  3. [Section 3.2 / Section 5.3] The phrase 'consistent with a theoretical prediction of about 48%–81% (Proposition 2)' overstates the status of the cost model: Propositions 1 and 2 are upper bounds under hand-chosen parameters (δ, λ, γ, κ), not predictions fitted to or tested against the latency measurements. Suggest rewording to 'consistent with the upper-bound analysis under the stated parameter regime'.
  4. [Limitations] The Limitations section discusses only the choice of base model and RL method; it does not mention the two main threats to the long-document results, namely the potential training/evaluation overlap and the teacher-dependence of the SFT trajectories. A sentence or two acknowledging these would be appropriate.
  5. [Section 6 / Conclusion] The first sentence of the Conclusion contains a typo: 'reducing, hallucination' should be 'reducing hallucination'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the training signal is teacher distillation, the theoretical bounds are not fitted to the measured results, and the headline accuracy claims are anchored to external benchmarks.

full rationale

The paper's central claim is an empirical one: SFT on 17.9K zoom-in trajectories plus GRPO on 19.2K hard examples improves Qwen3-VL-8B on DUDE, MP-DocVQA, MMLongBench-Doc, LongDocURL, MME-RealWorld-Lite, and O3-Bench (Tables 2, 3, 4, 16). The only self-citation that plays a substantive role is InSight-o3 (Li et al., 2026), used as the two-agent teacher that produces SFT trajectories (Section 4.2, Figure 3). This is distillation, not circular reasoning: the teacher supplies training data, not the evaluated quantity, and the trajectories are filtered by answer correctness and audited in Appendix D.2 (e.g., 95.33% evidence-page hit rate). The latency and sequence-length 'predictions' in Propositions 1 and 2 are derived upper bounds with declared parameters (delta, lambda, gamma, kappa); the paper reports measured latency separately and uses the propositions only as consistency checks, so they are not fits renamed as predictions. The paper also flags its own limitations: the Limitations section states only one base model was used and no advanced RL methods were tried; Appendix F.3 states cross-paper numbers are 'not fully controlled'; Appendix F.2 shows the 40-page cap hides part of the difficulty. One genuine validity risk is not a circularity: the arXiv training corpus is built by 'prioritising longer papers that are more likely to contain dense figures and tables' (Appendix C.1), while MMLongBench-Doc contains scientific papers up to 468 pages (Appendix E.1), and no deduplication against the evaluation benchmarks is reported. If actual PDFs overlapped, gains could be inflated by memorization, but the paper provides no evidence of overlap, and the hard rule here forbids speculative circularity. Under the stated criteria, no load-bearing step reduces to its own input.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central accuracy claim rests primarily on empirical training and evaluation, not on a mathematical derivation. The theoretical latency bounds use hand-chosen parameters (delta, lambda, gamma, kappa) and assume an approximate latency model. The main domain assumptions are that context rot exists, that low-resolution overviews retain navigational structure, and that the InSight-o3 teacher provides reliable zoom trajectories. No new physical or conceptual entities are introduced.

free parameters (3)
  • delta (relative input-token cost per tool call) = assumed range 0.01-0.05
    Used in the latency model (Eq. 3) to bound sequence length and latency reductions; not fitted to measurements, so the theoretical bounds are illustrative rather than predictive.
  • lambda (relative output-token cost per tool call) = assumed range 0.10-0.50
    Same role in Eq. 3; hand-chosen ranges make the latency upper bounds in Table 1 illustrative, not derivations of the measured 41-68% reduction.
  • gamma and kappa (cost ratio and prompt-to-response ratio) = gamma=0.01, kappa=100 (representative)
    Used in Proposition 2 to compute the weights in Eq. 6; representative values, not fitted to the actual model.
assumptions (5)
  • domain assumption Context rot exists and degrades performance as context length grows.
    The motivation relies on this (Hong et al. 2025, Anthropic 2025); if context rot were not real, the efficiency-accuracy tradeoff would look different.
  • domain assumption Inference latency follows T(P,R) = alpha P^2 + beta R(2P+R) with beta >= alpha and prefix caching.
    Used in Propositions 1-2; an approximation that may not hold for all backends.
  • domain assumption Low-resolution page images (50 DPI) retain enough structure for the agent to locate relevant regions.
    Central to the coarse-to-fine design; if the initial overview is too blurry, the agent cannot pick where to zoom.
  • domain assumption The InSight-o3 two-agent teacher produces correct, unbiased zoom trajectories.
    The SFT set is distilled from InSight-o3 (same group's prior work); teacher errors in bounding boxes or reasoning would propagate to the student.
  • standard math Standard inequalities and algebra in Appendix B are valid.
    The proofs use simple algebraic manipulation; no advanced machinery.

how reviews work

0 comments
Cite this review

Pith. "Pith review of InSight-doc: Agentic Visual Perception for Long-Document Understanding." pith.science (2026). https://pith.science/paper/XNCSBJ22

@misc{pith2026260810628,
  author       = {Pith},
  title        = {Pith review of: InSight-doc: Agentic Visual Perception for Long-Document Understanding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XNCSBJ22}},
  note         = {Machine review of arXiv:2608.10628}
}
read the original abstract

Long-document understanding often requires reasoning over many visually rich pages, making inference costly and prone to context rot. In this work, we propose InSight-doc, an agentic visual perception framework that treats visual resolution as an adaptive reasoning-time resource. InSight-doc starts from low resolution and selectively zooms into high-resolution regions for finer evidence, without relying on any external retriever. To train such an agent, we construct an active-perception corpus of 17.9K high-quality SFT examples with region-level zoom-in trajectories, accompanied by 19.2K hard RL examples. Through SFT+RL, InSight-doc-8B improves the baseline by 4.3--16.4 accuracy points over document VQA benchmarks. On long documents, it reduces hallucination by more than 40% and inference latency by 41%--68% while maintaining an accuracy lead. Our code, datasets, and model are released at https://github.com/m-Just/InSight-doc .

Figures

Figures reproduced from arXiv: 2608.10628 by the authors.

Figure 1
Figure 1. InSight-doc substantially reduces hallucina￾tion, sequence length, and latency while improving accuracy on long-document VQA. Hallucination rate is measured on unanswerable questions as the rate of non-abstaining answers. More details in Appendix F.1. answering questions over lengthy, visually rich doc￾uments such as research papers and financial re￾ports (Tito et al., 2023; Van Landeghem et al., 2023; Ma et al., 20… view at source ↗
Figure 2
Figure 2. Illustration of InSight-doc on a long-document VQA example. Starting from a low-resolution overview, InSight-doc performs an interleaved multimodal chain-of-thought: each round emits a thought (<think>) and a zoom-in tool call (<tool_call>) specifying img_idx, label, and bbox; the cropped high-resolution region is then appended as visual evidence. After multiple rounds of active perception, the model produces the fi… view at source ↗
Figure 3
Figure 3. Overview of data construction pipeline. The top panel shows a three-stage filtering and CoT construction process: Stage 1 discards questions answerable without the document, Stage 2 discards questions already answerable at low DPI without zoom, and Stage 3 uses InSight-o3 to construct zoom-in CoTs for the remaining items, routing successes to SFT and failures to RL. The bottom panel illustrates the InSight-o3 two-ag… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Accuracy vs. efficiency on four multi-page document VQA benchmarks. We compare InSight-doc (ours, green) with Qwen3-VL-8B without zoom-in calls (purple) and with zoom-in calls (pink). For each method, darker shades indicate larger input resize ratios (higher resolution…
Figure 5
Figure 5. Figure 5: Statistics of our training data. Since β ≥ α and all token counts are nonnega￾tive, Eq. (11) implies Tr ≤ T(Pr, Rr). Substituting Pr = x(r)P0 and Rr = y(r)R0 gives Tr ≤ αx(r) 2P 2 0 +βy(r)R0 (2x(r)P0 + y(r)R0). The no-resize baseline has latency T0 = αP2 0 + βR0(2P0 + …
Figure 6
Figure 6. Figure 6: An example of our SFT data. E.3 Judge Calibration Our reward model relies on an LLM judge to com￾pare the model’s final answer against the ground￾truth answer. We first used a legacy two-stage judge. This judge asks the LLM to extract a con￾cise final answer from the m…
Figure 7
Figure 7. Figure 7: Averaged over MMLongBench-Doc and LongDocURL, its gains over Qwen3-VL-8B on the longest-document subset are +11.6, +6.4, and +6.4 points at r = 0.25, 0.35, 0.5, respectively. Compared with the results on the full evalua￾tion set, the accuracy-efficiency tradeoff become…
Figure 8
Figure 8. Figure 8: Example of InSight-doc on an unanswerable question, i.e., one whose answer cannot be supported by any [PITH_FULL_IMAGE:figures/full_fig_p027_8.png]
Figure 9
Figure 9. Figure 9: Examples illustrating InSight-doc’s sequential zoom-in behavior, where it progressively refines target [PITH_FULL_IMAGE:figures/full_fig_p028_9.png]
Figure 10
Figure 10. Figure 10: An example demonstrating that InSight-doc can think and see in a manner similar to humans:adjusting [PITH_FULL_IMAGE:figures/full_fig_p029_10.png]
Figure 11
Figure 11. Figure 11: An example demonstrating that InSight-doc can think and see in a manner similar to humans: actively [PITH_FULL_IMAGE:figures/full_fig_p030_11.png]
Figure 12
Figure 12. Figure 12: Examples demonstrating that InSight-doc can gather evidence from different pages. [PITH_FULL_IMAGE:figures/full_fig_p031_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

199 extracted references · 93 linked inside Pith

  1. [1]

    arXiv preprint arXiv:2512.12658 , year=

    CogDoc: Towards Unified thinking in Documents , author=. arXiv preprint arXiv:2512.12658 , year=

    arXiv

  2. [2]

    2026 , url=

    Paper2Poster: Towards Multimodal Poster Automation from Scientific Papers , author=. 2026 , url=

    2026

  3. [3]

    arXiv:2602.18600 , year=

    MapTab: Can MLLMs Master Constrained Route Planning? , author=. arXiv:2602.18600 , year=

    arXiv

  4. [4]

    Gemini 3.1 Flash-Lite Preview , year =

  5. [5]

    arXiv preprint arXiv:2603.13398 , year=

    Qianfan-OCR: A Unified End-to-End Model for Document Intelligence , author=. arXiv preprint arXiv:2603.13398 , year=

    arXiv

  6. [6]

    DocSeeker: Structured Visual Reasoning with Evidence Grounding for Long Document Understanding , author =

  7. [7]

    arXiv preprint arXiv:2604.13731 , year=

    Doc-V*: Coarse-to-Fine Interactive Visual Reasoning for Multi-Page Document VQA , author=. arXiv preprint arXiv:2604.13731 , year=

    arXiv

  8. [8]

    2023 , publisher=

    Hierarchical multimodal transformers for multipage docvqa , author=. 2023 , publisher=

    2023

Show all 199 references
  1. [9]

    Document understanding dataset and evaluation (dude) , author=

  2. [10]

    mplug-docowl2: High-resolution compressing for ocr-free multi-page document understanding , author=

  3. [11]

    Monkey: Image resolution and text label are important things for large multi-modal models , author=

  4. [12]

    Mmlongbench-doc: Benchmarking long-context document understanding with visualizations , author=

  5. [13]

    Docvqa: A dataset for vqa on document images , author=

  6. [14]

    arXiv:2008.08899 , year =

    Document visual question answering challenge 2020 , author =. arXiv:2008.08899 , year =

    2020 arXiv

  7. [15]

    Layoutlm: Pre-training of text and layout for document image understanding , author=

  8. [16]

    Layoutlmv2: Multi-modal pre-training for visually-rich document understanding , author=

  9. [17]

    Layoutlmv3: Pre-training for document ai with unified text and image masking , author=

  10. [18]

    Visual instruction tuning , author=

  11. [19]

    arXiv:2308.12966 , year =

    Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond , author =. arXiv:2308.12966 , year =

    arXiv

  12. [20]

    arXiv preprint arXiv:2510.18234 , year=

    Deepseek-ocr: Contexts optical compression , author=. arXiv preprint arXiv:2510.18234 , year=

    arXiv

  13. [21]

    arXiv:2403.05530 , year =

    arXiv

  14. [22]

    arXiv:2304.14178 , year =

    mplug-owl: Modularization empowers large language models with multimodality , author =. arXiv:2304.14178 , year =

    arXiv

  15. [23]

    Cogagent: A visual language model for gui agents , author=

  16. [24]

    2024 , publisher=

    Docpedia: Unleashing the power of large multimodal model in the frequency domain for versatile document understanding , author=. 2024 , publisher=

    2024

  17. [25]

    arXiv:2410.21276 , year =

    Gpt-4o system card , author =. arXiv:2410.21276 , year =

    arXiv

  18. [26]

    arXiv:2312.11805 , year =

    arXiv

  19. [27]

    arXiv:2307.02499 , year =

    mplug-docowl: Modularized multimodal large language model for document understanding , author =. arXiv:2307.02499 , year =

    arXiv

  20. [28]

    arXiv:2308.13418 , year =

    Nougat: Neural optical understanding for academic documents , author =. arXiv:2308.13418 , year =

    arXiv

  21. [29]

    Retrieval-augmented generation for knowledge-intensive nlp tasks , author=

  22. [30]

    Leveraging passage retrieval with generative models for open domain question answering , author=

  23. [31]

    arXiv:2407.01449 , year =

    Colpali: Efficient document retrieval with vision language models , author =. arXiv:2407.01449 , year =

    arXiv

  24. [32]

    Lost in the middle: How language models use long contexts , author=

  25. [33]

    Chain-of-thought prompting elicits reasoning in large language models , author=

  26. [34]

    Large language models are zero-shot reasoners , author=

  27. [35]

    arXiv:2203.11171 , year =

    Self-consistency improves chain of thought reasoning in language models , author =. arXiv:2203.11171 , year =

    arXiv

  28. [36]

    Tree of thoughts: Deliberate problem solving with large language models , author=

  29. [37]

    Ureader: Universal ocr-free visually-situated language understanding with multimodal large language model , author=

  30. [38]

    arXiv:2403.04473 , year =

    Textmonkey: An ocr-free large multimodal model for understanding document , author =. arXiv:2403.04473 , year =

    arXiv

  31. [39]

    Longdocurl: a comprehensive multimodal long document benchmark integrating understanding, reasoning, and locating , author=

  32. [40]

    Slidevqa: A dataset for document visual question answering on multiple images , author=

  33. [41]

    Bert: Pre-training of deep bidirectional transformers for language understanding , author=

  34. [42]

    arXiv:2004.05150 , year =

    Longformer: The long-document transformer , author =. arXiv:2004.05150 , year =

    2004 arXiv

  35. [43]

    arXiv preprint arXiv:2511.15090 , year=

    BBox DocVQA: A Large Scale Bounding Box Grounded Dataset for Enhancing Reasoning in Document Visual Question Answer , author=. arXiv preprint arXiv:2511.15090 , year=

  36. [44]

    A Simple yet Effective Layout Token in Large Language Models for Document Understanding , author=

  37. [45]

    Docopilot: Improving Multimodal Models for Document-Level Understanding , author=

  38. [46]

    CREAM: coarse-to-fine retrieval and multi-modal efficient tuning for document VQA , author=

  39. [47]

    Docvlm: Make your vlm an efficient reader , author=

  40. [48]

    arXiv:2411.04952 , year =

    M3docrag: Multi-modal retrieval is what you need for multi-page multi-document understanding , author =. arXiv:2411.04952 , year =

    arXiv

  41. [49]

    Unifying Multimodal Retrieval via Document Screenshot Embedding , author=

  42. [50]

    Visdom: Multi-document qa with visually rich elements using multimodal retrieval-augmented generation , author=

  43. [51]

    arXiv:2508.07313 , year =

    Docr1: Evidence page-guided grpo for multi-page document understanding , author =. arXiv:2508.07313 , year =

  44. [52]

    arXiv:2205.10625 , year =

    Least-to-most prompting enables complex reasoning in large language models , author =. arXiv:2205.10625 , year =

    arXiv

  45. [53]

    arXiv:2410.10594 , year =

    Visrag: Vision-based retrieval-augmented generation on multi-modality documents , author =. arXiv:2410.10594 , year =

    arXiv

  46. [54]

    Vdocrag: Retrieval-augmented generation over visually-rich documents , author=

  47. [55]

    Doc-React: Multi-page Heterogeneous Document Question-answering , author=

  48. [56]

    arXiv:2308.11592 , year =

    Unidoc: A universal large multimodal model for simultaneous text detection, recognition, spotting and understanding , author =. arXiv:2308.11592 , year =

    arXiv

  49. [57]

    Efficient memory management for large language model serving with pagedattention , author=

  50. [58]

    arXiv:2411.01106 , year =

    SV-RAG: LoRA-Contextualizing Adaptation of MLLMs for Long Document Understanding , author =. arXiv:2411.01106 , year =

    arXiv

  51. [59]

    arXiv:2410.05970 , year =

    Wukong: A large multimodal model for efficient long pdf reading with end-to-end sparse sampling , author =. arXiv:2410.05970 , year =

    arXiv

  52. [60]

    Zou, Anni and Yu, Wenhao and Zhang, Hongming and Ma, Kaixin and Cai, Deng and Zhang, Zhuosheng and Zhao, Hai and Yu, Dong , booktitle = KNOWLEDGENLP, pages =

  53. [61]

    Dong, Kuicai and Chang, Yujing and Goh Xin Deik, Derrick and Li, Dexun and Tang, Ruiming and Liu, Yong , booktitle = EMNLP, pages =

  54. [62]

    Chia, Yew Ken and Cheng, Liying and Chan, Hou Pong and Song, Maojia and Liu, Chaoqun and Aljunied, Mahani and Poria, Soujanya and Bing, Lidong , booktitle = EMNLP, pages =

  55. [63]

    arXiv:2409.18839 , year =

    Mineru: An open-source solution for precise document content extraction , author =. arXiv:2409.18839 , year =

    arXiv

  56. [64]

    arXiv:2505.14059 , year =

    Dolphin: Document image parsing via heterogeneous anchor prompting , author =. arXiv:2505.14059 , year =

    arXiv

  57. [65]

    arXiv:2506.05218 , year =

    MonkeyOCR: Document Parsing with a Structure-Recognition-Relation Triplet Paradigm , author =. arXiv:2506.05218 , year =

  58. [66]

    2025 , publisher=

    Large language models in document intelligence: A comprehensive survey, recent advances, challenges and future trends , author=. 2025 , publisher=

    2025

  59. [67]

    2008 , publisher=

    Human reasoning and cognitive science , author=. 2008 , publisher=

    2008

  60. [68]

    Attention is all you need , author=

  61. [69]

    arXiv:2010.11929 , year =

    An image is worth 16x16 words: Transformers for image recognition at scale , author =. arXiv:2010.11929 , year =

    2010 arXiv

  62. [70]

    arXiv:2412.16720 , year =

    Openai o1 system card , author =. arXiv:2412.16720 , year =

    arXiv

  63. [71]

    2022 , organization=

    Ocr-free document understanding transformer , author=. 2022 , organization=

    2022

  64. [72]

    2023 , organization=

    Pix2struct: Screenshot parsing as pretraining for visual language understanding , author=. 2023 , organization=

    2023

  65. [74]

    arXiv:2412.10302 , year =

    DeepSeek-VL2: Mixture-of-Experts Vision-Language Models for Advanced Multimodal Understanding , author =. arXiv:2412.10302 , year =

    arXiv

  66. [75]

    arXiv:2408.01800 , year =

    MiniCPM-V: A GPT-4V Level MLLM on Your Phone , author =. arXiv:2408.01800 , year =

    arXiv

  67. [76]

    Li, Linjie and Wang, Yuxuan and Xu, Rui and Wang, Peiyi and Feng, Xinyun and Kong, Lingpeng and Liu, Qun , booktitle = ACL, pages =

  68. [77]

    Ding, Yihao and Ren, Kaixuan and Huang, Jiabin and Luo, Siwen and Han, Soyeon Caren , booktitle =

  69. [78]

    2007 , publisher=

    A normalized Levenshtein distance metric , author=. 2007 , publisher=

    2007

  70. [79]

    A brief analysis of ChatGPT: historical evolution current applications and future prospects [J] , author=

  71. [80]

    Dongsheng Wang and Natraj Raman and Mathieu Sibue and Zhiqiang Ma and Petr Babkin and Simerjot Kaur and Yulong Pei and Armineh Nourbakhsh and Xiaomo Liu , booktitle =

  72. [81]

    Shitao Xiao and Zheng Liu and Peitian Zhang and Niklas Muennighoff , booktitle =

  73. [82]

    arXiv:2507.06261 , year=

    Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities , author=. arXiv:2507.06261 , year=

    arXiv

  74. [83]

    arXiv:2408.03326 , year=

    Llava-onevision: Easy visual task transfer , author=. arXiv:2408.03326 , year=

    arXiv

  75. [84]

    V*: Guided visual search as a core mechanism in multimodal llms , author=

  76. [85]

    arXiv:2408.13257 , year=

    Mme-realworld: Could your multimodal llm challenge high-resolution real-world scenarios that are difficult for humans? , author=. arXiv:2408.13257 , year=

    arXiv

  77. [86]

    arXiv:2411.16044 , year=

    Zoomeye: Enhancing multimodal llms with human-like zooming capabilities through tree-based image exploration , author=. arXiv:2411.16044 , year=

    arXiv

  78. [87]

    Dyfo: A training-free dynamic focus visual search for enhancing lmms in fine-grained visual understanding , author=

  79. [88]

    arXiv:2505.15966 , year=

    Pixel reasoner: Incentivizing pixel-space reasoning with curiosity-driven reinforcement learning , author=. arXiv:2505.15966 , year=

    arXiv

  80. [89]

    Thinking with Images

    DeepEyes: Incentivizing" Thinking with Images" via Reinforcement Learning , author=. arXiv:2505.14362 , year=

    arXiv

  81. [90]

    arXiv:2505.21457 , year=

    Active-O3: Empowering Multimodal Large Language Models with Active Perception via GRPO , author=. arXiv:2505.21457 , year=

    arXiv

  82. [91]

    arXiv:2506.11991 , year=

    Vgr: Visual grounded reasoning , author=. arXiv:2506.11991 , year=

    arXiv

  83. [92]

    arXiv:2507.07999 , year=

    Traceable evidence enhanced visual grounded reasoning: Evaluation and methodology , author=. arXiv:2507.07999 , year=

  84. [93]

    arXiv:2509.07969 , year=

    Mini-o3: Scaling Up Reasoning Patterns and Interaction Turns for Visual Search , author=. arXiv:2509.07969 , year=

    arXiv

  85. [94]

    Mmmu: A massive multi-discipline multimodal understanding and reasoning benchmark for expert agi , author=

  86. [95]

    arXiv:2505.21327 , year=

    MME-Reasoning: A Comprehensive Benchmark for Logical Reasoning in MLLMs , author=. arXiv:2505.21327 , year=

    arXiv

  87. [96]

    Can MLLMs Reason in Multimodality? EMMA: An Enhanced MultiModal ReAsoning Benchmark , author=

  88. [97]

    Divide, conquer and combine: A training-free framework for high-resolution image perception in multimodal large language models , author=

  89. [98]

    Segment anything , author=

  90. [99]

    Ntire 2017 challenge on single image super-resolution: Dataset and study , author=

    2017

  91. [100]

    2020 , publisher=

    A comprehensive benchmark for single image compression artifact reduction , author=. 2020 , publisher=

    2020

  92. [101]

    Benchmarking ultra-high-definition image super-resolution , author=

  93. [102]

    arXiv:2306.05390 , year=

    Hq-50k: A large-scale, high-quality dataset for image restoration , author=. arXiv:2306.05390 , year=

    arXiv

  94. [103]

    ISPRS Journal of Photogrammetry and Remote Sensing , volume=

    FAIR1M: A benchmark dataset for fine-grained object recognition in high-resolution remote sensing imagery , author=. ISPRS Journal of Photogrammetry and Remote Sensing , volume=. 2022 , publisher=

    2022

  95. [104]

    2022 , organization=

    Coda: A real-world road corner case dataset for object detection in autonomous driving , author=. 2022 , organization=

    2022

  96. [105]

    Rank2tell: A multimodal driving dataset for joint importance ranking and reasoning , author=

  97. [106]

    2021 , publisher=

    Detection and tracking meet drones challenge , author=. 2021 , publisher=

    2021

  98. [107]

    LLVIP: A visible-infrared paired dataset for low-light vision , author=

  99. [108]

    arXiv:2402.03300 , year=

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models , author=. arXiv:2402.03300 , year=

    arXiv

  100. [109]

    arXiv:2501.12948 , year=

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning , author=. arXiv:2501.12948 , year=

    arXiv

  101. [110]

    arXiv:2503.14476 , year=

    Dapo: An open-source llm reinforcement learning system at scale , author=. arXiv:2503.14476 , year=

    arXiv

  102. [111]

    Machine learning , volume=

    Simple statistical gradient-following algorithms for connectionist reinforcement learning , author=. Machine learning , volume=. 1992 , publisher=

    1992

  103. [112]

    arXiv:2501.03262 , year=

    Reinforce++: An efficient rlhf algorithm with robustness to both prompt and reward models , author=. arXiv:2501.03262 , year=

    arXiv

  104. [113]

    arXiv:1707.06347 , year=

    Proximal policy optimization algorithms , author=. arXiv:1707.06347 , year=

    arXiv

  105. [114]

    Direct preference optimization: Your language model is secretly a reward model , author=

  106. [115]

    arXiv:2503.06749 , year=

    Vision-r1: Incentivizing reasoning capability in multimodal large language models , author=. arXiv:2503.06749 , year=

    arXiv

  107. [116]

    arXiv:2503.10615 , year=

    R1-onevision: Advancing generalized multimodal reasoning through cross-modal formalization , author=. arXiv:2503.10615 , year=

    arXiv

  108. [117]

    arXiv:2503.07365 , year=

    Mm-eureka: Exploring the frontiers of multimodal reasoning with rule-based reinforcement learning , author=. arXiv:2503.07365 , year=

    arXiv

  109. [118]

    arXiv:2504.11468 , year=

    Sft or rl? an early investigation into training r1-like reasoning large vision-language models , author=. arXiv:2504.11468 , year=

    arXiv

  110. [119]

    arXiv:2504.08837 , year=

    Vl-rethinker: Incentivizing self-reflection of vision-language models with reinforcement learning , author=. arXiv:2504.08837 , year=

    arXiv

  111. [120]

    arXiv:2504.07615 , year=

    Vlm-r1: A stable and generalizable r1-style large vision-language model , author=. arXiv:2504.07615 , year=

    arXiv

  112. [121]

    arXiv:2506.04207 , year=

    Advancing Multimodal Reasoning: From Optimized Cold Start to Staged Reinforcement Learning , author=. arXiv:2506.04207 , year=

  113. [122]

    arXiv:2503.17352 , year=

    OpenVLThinker: An Early Exploration to Complex Vision-Language Reasoning via Iterative Self-Improvement , author=. arXiv:2503.17352 , year=

    arXiv

  114. [123]

    arXiv:2507.05255 , year=

    Open vision reasoner: Transferring linguistic cognitive behavior for visual reasoning , author=. arXiv:2507.05255 , year=

  115. [124]

    arXiv:2504.05599 , year=

    Skywork r1v: Pioneering multimodal reasoning with chain-of-thought , author=. arXiv:2504.05599 , year=

    arXiv

  116. [125]

    arXiv:2504.16656 , year=

    Skywork r1v2: Multimodal hybrid reinforcement learning for reasoning , author=. arXiv:2504.16656 , year=

    arXiv

  117. [126]

    arXiv:2507.06167 , year=

    Skywork-r1v3 technical report , author=. arXiv:2507.06167 , year=

    arXiv

  118. [127]

    arXiv:2504.10479 , year=

    Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models , author=. arXiv:2504.10479 , year=

    arXiv

  119. [128]

    arXiv:2507.01949 , year=

    Kwai Keye-VL Technical Report , author=. arXiv:2507.01949 , year=

    arXiv

  120. [129]

    arXiv:2509.01563 , year=

    Kwai Keye-VL 1.5 Technical Report , author=. arXiv:2509.01563 , year=

    arXiv

  121. [130]

    arXiv:2502.13923 , year=

    Qwen2.5-vl technical report , author=. arXiv:2502.13923 , year=

    arXiv

  122. [131]

    5-vl technical report , author=

    Seed1. 5-vl technical report , author=. arXiv:2505.07062 , year=

    arXiv

  123. [132]

    arXiv:2504.07491 , year=

    Kimi-vl technical report , author=. arXiv:2504.07491 , year=

    arXiv

  124. [133]

    arXiv:2506.03569 , year=

    MiMo-VL Technical Report , author=. arXiv:2506.03569 , year=

    arXiv

  125. [134]

    arXiv:2507.01006 , year=

    GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning , author=. arXiv:2507.01006 , year=

    arXiv

  126. [135]

    CoRR , year=

    Visual cot: Unleashing chain-of-thought reasoning in multi-modal language models , author=. CoRR , year=

  127. [136]

    arXiv:2402.04236 , year=

    Cogcom: Train large vision-language models diving into details through chain of manipulations , author=. arXiv:2402.04236 , year=

    arXiv

  128. [137]

    Visual sketchpad: Sketching as a visual chain of thought for multimodal language models , author=

  129. [138]

    arXiv:2507.07998 , year=

    Pyvision: Agentic vision with dynamic tooling , author=. arXiv:2507.07998 , year=

    arXiv

  130. [139]

    arXiv:2508.11630 , year=

    Thyme: Think Beyond Images , author=. arXiv:2508.11630 , year=

    arXiv

  131. [140]

    arXiv:2505.14246 , year=

    Visual Agentic Reinforcement Fine-Tuning , author=. arXiv:2505.14246 , year=

    arXiv

  132. [141]

    arXiv:2505.15436 , year=

    Chain-of-Focus: Adaptive Visual Search and Zooming for Multimodal Reasoning via RL , author=. arXiv:2505.15436 , year=

    arXiv

  133. [142]

    arXiv:2505.15879 , year=

    GRIT: Teaching MLLMs to Think with Images , author=. arXiv:2505.15879 , year=

    arXiv

  134. [143]

    arXiv:2505.08617 , year=

    Openthinkimg: Learning to think with images via visual tool reinforcement learning , author=. arXiv:2505.08617 , year=

    arXiv

  135. [144]

    arXiv:2505.19702 , year=

    Point-rft: Improving multimodal reasoning with visually grounded reinforcement finetuning , author=. arXiv:2505.19702 , year=

    arXiv

  136. [145]

    arXiv:2505.15510 , year=

    Visual thoughts: A unified perspective of understanding multimodal chain-of-thought , author=. arXiv:2505.15510 , year=

  137. [146]

    arXiv:2505.16192 , year=

    VLM-R3: Region Recognition, Reasoning, and Refinement for Enhanced Multimodal Chain-of-Thought , author=. arXiv:2505.16192 , year=

    arXiv

  138. [147]

    arXiv:2506.04034 , year=

    Rex-Thinker: Grounded Object Referring via Chain-of-Thought Reasoning , author=. arXiv:2506.04034 , year=

    arXiv

  139. [148]

    arXiv:2508.12109 , year=

    Simple o3: Towards Interleaved Vision-Language Reasoning , author=. arXiv:2508.12109 , year=

    arXiv

  140. [149]

    Making the v in vqa matter: Elevating the role of image understanding in visual question answering , author=

  141. [150]

    Gqa: A new dataset for real-world visual reasoning and compositional question answering , author=

  142. [151]

    Vizwiz grand challenge: Answering visual questions from blind people , author=

  143. [152]

    International Journal on Digital Libraries , volume=

    Scienceqa: A novel resource for question answering on scholarly articles , author=. International Journal on Digital Libraries , volume=. 2022 , publisher=

    2022

  144. [153]

    arXiv:2305.10355 , year=

    Evaluating object hallucination in large vision-language models , author=. arXiv:2305.10355 , year=

    arXiv

  145. [154]

    arXiv:2306.13394 , year=

    MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models , author=. arXiv:2306.13394 , year=

    arXiv

  146. [155]

    arXiv:2307.06281 , year=

    MMBench: Is Your Multi-modal Model an All-around Player? , author=. arXiv:2307.06281 , year=

    arXiv

  147. [156]

    arXiv:2404.14396 , year=

    Seed-x: Multimodal models with unified multi-granularity comprehension and generation , author=. arXiv:2404.14396 , year=

    arXiv

  148. [157]

    arXiv:2310.02255 , year=

    Mathvista: Evaluating mathematical reasoning of foundation models in visual contexts , author=. arXiv:2310.02255 , year=

    arXiv

  149. [158]

    Measuring multimodal mathematical reasoning with math-vision dataset , author=

  150. [159]

    2024 , organization=

    Mathverse: Does your multi-modal llm truly see the diagrams in visual math problems? , author=. 2024 , organization=

    2024

  151. [160]

    arXiv:2409.02813 , year=

    Mmmu-pro: A more robust multi-discipline multimodal understanding benchmark , author=. arXiv:2409.02813 , year=

    arXiv

  152. [161]

    arXiv:2507.05595 , year=

    PaddleOCR 3.0 Technical Report , author=. arXiv:2507.05595 , year=

    arXiv

  153. [162]

    arXiv:2409.19256 , year =

    HybridFlow: A Flexible and Efficient RLHF Framework , author =. arXiv:2409.19256 , year =

    arXiv

  154. [163]

    Infographicvqa , author=

  155. [164]

    arXiv:2503.20020 , year=

    Gemini robotics: Bringing ai into the physical world , author=. arXiv:2503.20020 , year=

    arXiv

  156. [165]

    arXiv:2506.04308 , year=

    RoboRefer: Towards Spatial Referring with Reasoning in Vision-Language Models for Robotics , author=. arXiv:2506.04308 , year=

  157. [166]

    arXiv:2407.01509 , year=

    MIA-Bench: Towards Better Instruction Following Evaluation of Multimodal LLMs , author=. arXiv:2407.01509 , year=

    arXiv

  158. [167]

    arXiv:2404.07972 , year=

    OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments , author=. arXiv:2404.07972 , year=

    arXiv

  159. [168]

    arXiv:2405.14573 , year=

    AndroidWorld: A Dynamic Benchmarking Environment for Autonomous Agents , author=. arXiv:2405.14573 , year=

    arXiv

  160. [169]

    Visual programming: Compositional visual reasoning without training , author=

  161. [170]

    Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face , author=

  162. [171]

    Vipergpt: Visual inference via python execution for reasoning , author=

  163. [172]

    2024 , organization=

    Hydra: A hyper agent for dynamic compositional visual reasoning , author=. 2024 , organization=

    2024

  164. [173]

    Feudal reinforcement learning , author=

  165. [174]

    Socratic Models: Composing Zero-Shot Multimodal Reasoning with Language , author=

  166. [175]

    NeurIPS 2024 Workshop on Compositional Learning: Perspectives, Methods, and Paths Forward , year=

    HAMMR: HierArchical MultiModal React agents for generic VQA , author=. NeurIPS 2024 Workshop on Compositional Learning: Perspectives, Methods, and Paths Forward , year=

    2024

  167. [176]

    Camel: Communicative agents for" mind" exploration of large language model society , author=

  168. [177]

    MetaGPT: Meta programming for a multi-agent collaborative framework , author=

  169. [178]

    First Conference on Language Modeling , year=

    Autogen: Enabling next-gen LLM applications via multi-agent conversations , author=. First Conference on Language Modeling , year=

  170. [179]

    arXiv:2505.13516 , year=

    HALO: Hierarchical Autonomous Logic-Oriented Orchestration for Multi-Agent LLM Systems , author=. arXiv:2505.13516 , year=

    arXiv

  171. [180]

    arXiv:2505.19591 , year=

    Multi-Agent Collaboration via Evolving Orchestration , author=. arXiv:2505.19591 , year=

  172. [181]

    arXiv preprint arXiv:2602.20913 , year=

    LongVideo-R1: Smart Navigation for Low-cost Long Video Understanding , author=. arXiv preprint arXiv:2602.20913 , year=

    arXiv

  173. [182]

    arXiv e-prints , pages=

    Videodeepresearch: Long video understanding with agentic tool using , author=. arXiv e-prints , pages=

  174. [183]

    arXiv:2604.02891 , year =

    Progressive Video Condensation with MLLM Agent for Long-form Video Understanding , author =. arXiv:2604.02891 , year =

    arXiv

  175. [184]

    arXiv preprint arXiv:2512.24601 , year=

    Recursive language models , author=. arXiv preprint arXiv:2512.24601 , year=

    arXiv

  176. [185]

    arXiv:2506.12508 , year=

    Agentorchestra: A hierarchical multi-agent framework for general-purpose task solving , author=. arXiv:2506.12508 , year=

    arXiv

  177. [186]

    arXiv:2511.21631 , year=

    Qwen3-VL Technical Report , author=. arXiv:2511.21631 , year=

    arXiv

  178. [187]

    arXiv:2508.18265 , year=

    InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency , author=. arXiv:2508.18265 , year=

    arXiv

  179. [188]

    Vrag-rl: Empower vision-perception-based rag for visually rich information understanding via iterative reasoning with reinforcement learning , author=

  180. [189]

    2026 , url=

    InSight-o3: Empowering Multimodal Foundation Models with Generalized Visual Search , author=. 2026 , url=

    2026

  181. [190]

    2026 , url=

    GPT-5 System Card , author=. 2026 , url=

    2026

  182. [191]

    2025 , month =

    Context Rot: How Increasing Input Tokens Impacts LLM Performance , author =. 2025 , month =

    2025

  183. [192]

    2025 , month =

    Anthropic , title =. 2025 , month =

    2025

  184. [193]

    Halle-switch: Rethinking and controlling object existence hallucinations in large vision-language models for detailed caption , author=

  185. [194]

    Advances in Neural Information Processing Systems , volume=

    Seeing is Believing? Mitigating OCR Hallucinations in Multimodal Large Language Models , author=. Advances in Neural Information Processing Systems , volume=

  186. [195]

    arXiv preprint arXiv:2210.03629 , year=

    React: Synergizing reasoning and acting in language models , author=. arXiv preprint arXiv:2210.03629 , year=

    arXiv

  187. [196]

    Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    Molorag: Bootstrapping document understanding via multi-modal logic-aware retrieval , author=. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    2025

  188. [197]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    URaG: Unified retrieval and generation in multimodal LLMs for efficient long document understanding , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  189. [198]

    Findings of the Association for Computational Linguistics: ACL 2026 , pages=

    MM-Doc-R1: Training Agents for Long Document Visual Question Answering through Multi-turn Reinforcement Learning , author=. Findings of the Association for Computational Linguistics: ACL 2026 , pages=

    2026

  190. [199]

    arXiv preprint arXiv:2505.09388 , year=

    Qwen3 technical report , author=. arXiv preprint arXiv:2505.09388 , year=

    arXiv

  191. [200]

    Findings of the Association for Computational Linguistics: EMNLP 2023 , pages=

    A lightweight method to generate unanswerable questions in English , author=. Findings of the Association for Computational Linguistics: EMNLP 2023 , pages=

    2023

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.